Pedestrian dead reckoning (PDR) is one key localization technique using inertial momentum units (IMU) installed on a smartphone. However, IMU measurements are affected by the user’s unpredictable smartphone-carrying pattern, degrading the resulting localization accuracy and making its practical usage questionable. This article aims to tackle the issue by proposing a novel technique called WAveform GuIded Transformation of IMU measurements (WAIT), which transforms real IMU measurements to error-free waveforms. WAIT is based on a Resnet-based autoencoder architecture that is trained to generate the waveforms no matter what smartphone-carrying pattern the IMU measurement comes in. The resulting consistent waveform can be used as input for various elementary algorithms required for updating a user’s location and pedestrian trajectory, such as step count and heading direction estimation algorithms. Besides, WAIT’s architecture remains unchanged regardless of the number of concerned smartphone-carrying patterns, and there is no performance degradation. The effectiveness of WAIT is well verified by field experiments conducted with three different smartphone-carrying patterns and two users, showing that the average positioning error is reduced from 0.967 to 0.433 (m) compared to benchmark PDR algorithm using a different deep neural network (DNN) model for each smartphone-carrying pattern.